File size: 5,870 Bytes
33bd5b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d42b8f9
 
 
 
33bd5b5
 
 
d42b8f9
 
 
 
 
 
 
 
 
 
 
33bd5b5
d42b8f9
 
 
 
 
 
 
 
 
 
 
 
 
33bd5b5
 
d42b8f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33bd5b5
 
d42b8f9
 
 
 
 
 
 
 
 
 
 
 
 
 
33bd5b5
2fbe51b
d42b8f9
33bd5b5
d42b8f9
33bd5b5
d42b8f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
"""PMC-OA Dataset"""

import os
import jsonlines

import datasets

logger = datasets.logging.get_logger(__name__)

_CITATION = """\
@article{lin2023pmc,
  title={PMC-CLIP: Contrastive Language-Image Pre-training using Biomedical Documents},
  author={Lin, Weixiong and Zhao, Ziheng and Zhang, Xiaoman and Wu, Chaoyi and Zhang, Ya and Wang, Yanfeng and Xie, Weidi},
  journal={arXiv preprint arXiv:2303.07240},
  year={2023}
}
"""

_DESCRIPTION = """\
Foundation models trained on large-scale dataset gain a recent surge in CV and NLP. In contrast, development in biomedical domain lags far behind due to data scarcity.
To address this issue, we build and release PMC-OA, a biomedical dataset with 1.6M image-caption pairs collected from PubMedCentral's OpenAccess subset, which is 8 times larger than before.
PMC-OA covers diverse modalities or diseases, with majority of the image-caption samples aligned at finer-grained level, i.e., subfigure and subcaption.
While pretraining a CLIP-style model on PMC-OA, our model named PMC-CLIP achieves state-of-the-art results on various downstream tasks,
including image-text retrieval on ROCO, MedMNIST image classification, Medical VQA, i.e. +8.1% R@10 on image-text retrieval, +3.9% accuracy on image classification.
"""

_HOMEPAGE = "https://weixionglin.github.io/PMC-CLIP/"

_URLs = {
    "images": "https://huggingface.co/datasets/axiong/pmc_oa/resolve/main/images.zip",
    "pmc_oa_beta": "https://huggingface.co/datasets/axiong/pmc_oa/resolve/main/pmc_oa_beta.jsonl",
    "pmc_oa": "https://huggingface.co/datasets/axiong/pmc_oa/resolve/main/pmc_oa.jsonl",
}


class PMC_OA_Config(datasets.BuilderConfig):
    """BuilderConfig for PMC_OA"""

    def __init__(self, **kwargs):
        """
        Args:
            **kwargs: keyword arguments forwarded to super.
        """
        super(PMC_OA_Config, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)


class PMC_OA(datasets.GeneratorBasedBuilder):
    """PMC_OA Dataset"""

    VERSION = datasets.Version("1.0.0")
    BUILDER_CONFIGS = [
        PMC_OA_Config(
            name="pmc_oa_beta",
            description="<subfigure, caption> pairs. Subfigures detected by a DETR model.",
        ),
        PMC_OA_Config(
            name="pmc_oa",
            description="<subfigure, subcaption> pairs. Subfigures detected by a DETR model. Subcaptions detected by ChatGPT and aligned with subfigures.",
        ),
    ]

    def _info(self):
        if self.config.name == "pmc_oa_beta":
            return datasets.DatasetInfo(
                description=_DESCRIPTION,
                features=datasets.Features(
                    {
                        "image": datasets.Value("string"),
                        "caption": datasets.Value("string"),
                    }
                ),
                supervised_keys=None,
                citation=_CITATION,
                homepage=_HOMEPAGE,
            )
        elif self.config.name == "pmc_oa":
            return datasets.DatasetInfo(
                description=_DESCRIPTION,
                features=datasets.Features(
                    {
                        "image": datasets.Value("string"),
                        "caption": datasets.Value("string"),
                        "alignment_type": datasets.Value("string"),
                        "alignment_score": datasets.Value("float"),
                    }
                ),
                supervised_keys=None,
                citation=_CITATION,
                homepage=_HOMEPAGE,
            )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        downloaded_files = dl_manager.download_and_extract(_URLs)
        if self.config.name == "pmc_oa_beta":
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["pmc_oa_beta"], "image_dir": downloaded_files['images']}
                )
            ]
        elif self.config.name == "pmc_oa":
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["pmc_oa"], "image_dir": downloaded_files['images']}
                )
            ]

    def _generate_examples(self, filepath, image_dir):
        """Yields examples."""
        logger.info("generating examples from = %s", filepath)
         
        with jsonlines.open(filepath) as reader:
            for _id, obj in enumerate(reader):
                if self.config.name == "pmc_oa_beta":
                    relative_image_path = obj['image']
                    image_path = os.path.join(image_dir, "caption_T060_filtered_top4_sep_v0_subfigures", relative_image_path)
                    caption = obj['caption']
                    yield _id, {
                        "image": {
                            "path": image_path,
                            "bytes": open(image_path, "rb").read(),
                        },
                        "caption": caption,
                    }
                elif self.config.name == "pmc_oa":
                    relative_image_path = obj['image']
                    image_path = os.path.join(image_dir, "caption_T060_filtered_top4_sep_v0_subfigures", relative_image_path)
                    caption = obj['caption']
                    alignment_type = obj['alignment_type']
                    alignment_score = obj['alignment_score']
                    yield _id, {
                        "image": {
                            "path": image_path,
                            "bytes": open(image_path, "rb").read(),
                        },
                        "caption": caption,
                        "alignment_type": alignment_type,
                        "alignment_score": alignment_score,
                    }